AMD Instinct MI355X
gfx950
Records held
3
Runs
2,691
Operations
32,691 implementations
Records held on this GPU
Operation · workload
History
Record
Implementation
Since
Coverage by operation family
Family
Operations
Runs
With source
gemm
1
1,143
588
moe
1
782
466
mla-attention
1
766
412
Serving evidence
End-to-end serving results measured on this hardware — a separate corpus with its own resolver, never ranked against the kernel records above.
Model · workload
Scenario
Best reported
Runs
Llama-2-70B (MLPerf reference)llama2-70b-99 · Offline
offline
103,480 tok/s
7
Llama-2-70B (MLPerf reference)llama2-70b-99 · Offline
offline
103,480 tok/s
7
Llama-2-70B (MLPerf reference)llama2-70b-99 · Server
server
100,282 tok/s
7
Llama-2-70B (MLPerf reference)llama2-70b-99 · Server
server
100,282 tok/s
7
Llama-2-70B (MLPerf reference)llama2-70b-99 · Interactive
interactive
73,608 tok/s
6
Llama-2-70B (MLPerf reference)llama2-70b-99 · Interactive
interactive
73,608 tok/s
6
GPT-OSS-120B (MLPerf reference)gpt-oss-120b · Offline
offline
95,004 tok/s
5
GPT-OSS-120B (MLPerf reference)gpt-oss-120b · Server
server
82,136 tok/s
5
66 serving runs across 24 comparison groups · Open in the serving resolver →
Record activity
Nov
Dec
Jan
Feb
44Mar
9Apr
May
Jun
Jul
Aug
Sep
Oct
Sources: GPU MODE KernelBot (2026-04-07) · June 9 Researcher Reciprocity License v1.0last observed 2026-04-07